In conventional speaker diarization, homogeneous segmentation of an audio stream pursuant to the identity of a speaker may improve readability of an automatic transcription by organizing the audio stream into speaker turns. Speaker diarization may be applicable and improve utility of all kinds of voice communication from broadcasts, entertainment, meeting recordings, voice mail, etc.